Does It Pay Off to Include Non-Internet Households in an Internet Panel?

TL;DR: In this paper, the authors investigated whether it is possible to improve the representativeness of an Internet panel by including non-Internet households, i.e., households with no Internet access at the time of sampling, with cost-free equipment and an Internet connection.

Abstract: This paper investigates whether it is possible to improve the representativeness of an Internet panel by including non-Internet households. We study the LISS panel, managed by CentERdata, an Internet panel based on a probability sample that comprises approximately 5000 households. The LISS panel provides non-Internet households, households with no Internet access at the time of the sampling, with cost-free equipment and an Internet connection. Early 2010 the LISS panel contained 545 non-Internet households, this equals approximately 10% of the entire panel. The analyses show that particularly older households, non-western immigrants, and one-person households are less likely to have Internet access. The LISS panel includes a representative sample of non-Internet households except for households with high average age ("the oldest old"). Non-Internet households who participate in the panel show higher response rates on the individual questionnaires and lower attrition rates. While significant differences between the panel and the Dutch population remain, the complete LISS panel, with both Internet and non-Internet households, appears to be closer to the Dutch population than the panel consisting only of Internet households for all socio-demographic variables we tested. Furthermore, about half of the non-Internet households start to use the Internet after they have become panel members. They use less of the options offered by Internet, and mainly use the simpler applications, such as e-mail and information search, compared to persons living in Internet households. In this sense, they remain different from the original Internet households and continue to contribute to the quality of the panel data.